On 01-07-2020 18:10, Eric Sunshine wrote: > On Wed, Jul 1, 2020 at 5:46 AM Ben Wijen <ben@xxxxxxxxx> wrote: > > "git init --separate-git-dir=" checks only whether the path exists, > and errors out if it does; it doesn't care whether the directory is > empty or not. I'm wondering, therefore, if this check should be > tightened to more closely align the behavior of the two commands. > (Using the tighter semantic now doesn't prohibit loosening it in the > future, whereas it's harder to tighten behavior which started out > loose.) > About `git init`, I did take a look at it, but saw 'exist_ok' is always true when called from 'cmd_init_db' (see builtin/init_db.c:654) This means `git init` always allows when git_dir/real_git_dir already exists, which I understand for the worktree, but I doubt if one wants this for the repository. (when --separate-git-dir is used) Also, because 'exist_ok' is always true for `git init` and - with this patch - `git clone` checks the git_dir/real_git_dir before the 'init_db' call I'm wondering if the code in 'init_db' (builtin/init-db.c:394-400) can be removed. Then, even if we do take that route (no 'is_empty_dir', only 'path_exists') we must also address junk_git_dir_flags, as REMOVE_DIR_KEEP_TOPLEVEL would invalidate a second git clone. So, as this patch only fixes the problem at hand, IMHO separate patch-sets should be created for mentioned issues. Ben...
